摘要
选取镦粗缸活塞运动速度、砧子缸活塞运动速度、镦粗压力和加热时间作为输入参数,加热电流作为输出参数,并用数学模型、BP神经网络、加法网络、乘法网络以及神经网络与机理模型综合集成的五种方案来对加热电流进行预报。比较结果表明,综合集成模型将数学模型的知识集成到网络结构中,在“小样本”时,不仅能减少连接权值,而且能加快训练速度,提高泛化能力,在电镦机加热电流的预报中取得了良好效果。
The prediction model are established by employing such five projects as mathematic model, BP NN, BP NN ＋ mathematic model, BP NN ＊ mathematic model, synthetic method of neural network and mechanistic model. Velocities of upsetting cylinder and anvil cylinder,pressure of upsetting cylinder, heating time used as input parameters, heating current was as target parameter. The results of the research could be shown that a synthetic method of neural network and mechanistic model integrates the mechanistic model knowledge into the neural network structure, which can reduce the number of network weight, improved the learning speed,as well as generalization performance of the network at “lack of learning sample”. And the super effect could be achieved if it is employed to predict the heating current of electric upsetting.
出处
《现代制造工程》
CSCD
2005年第9期62-64,共3页
Modern Manufacturing Engineering
基金
广东省自然科学基金资助项目(990141)
关键词
神经网络
综合集成神经网络
加热电流
预报
Neural network Synthetic neural network Heating current Prediction
